New Kid on the Block: Here’s how Amsterdam AI startup Langwatch saves businesses from PR disasters and costly legal cases

In the rapidly evolving landscape of artificial intelligence, generative AI has emerged as a groundbreaking technology, empowering businesses to create innovative solutions and unlock new possibilities. 

However, as with any disruptive innovation, the integration of generative AI into products and services carries its inherent risks, from unintended biases to potential data leaks and reputational damage. 

Recognising this need, an Amsterdam-based startup, Langwatch, has stepped forward to address these challenges head-on, offering a quality and analytics platform designed to safeguard the responsible deployment of generative AI solutions.

As part of our ongoing series “New Kids on the Block,” we at Silicon Canals interviewed Manouk Draisma, CEO and co-founder of Langwatch.

In this interview, we delved into the need for Generative AI, its inherent risks, the company’s quality assurance techniques, and much more. 

Do give it a read. 

Birth of Langwatch: Identifying the need

The birth of Langwatch goes back to the founders’ personal experiences and observations within the AI industry. 

“While everyone is intrigued by the capabilities of the newest GenAI generation, I also noticed that my friends at enterprise companies were struggling to use it safely,” says Manouk Draisma. 

Rogerio Chaves and Manouk Draisma, the founders of Langwatch, met during an Antler residency in Amsterdam. They have more than 25 years of experience in the software industry, having worked at companies like Booking.com and Lightspeed. 

“When I met my co-founder, Rogerio, the problem he saw was similar to the one he saw when working at Booking.com, where they had limited control and insights into how users were using the product,” continues Draisma. 

“A mutual friend brought us together while both were working on GenAI products as a side-project. We figured out the real problem and need for a Quality Control & analytics tool, which has been presented to the birth of Langwatch,” explains Draisma. 

Rogerio brings extensive expertise in engineering and product development to the Langwatch team. On the other hand, Manouk brings a ton of experience on the commercial side, leading and building teams at both startups and public companies.

Mission and Vision

With companies around the world investing in new AI-powered tools, businesses are at risk of misusing and abusing AI. 

Misuse of AI tools, such as swearing chatbots and manipulation of purchases for $1, highlights the need for more control to protect brand reputation.

“Companies of all sizes are investing in building new tools or improving their current tool stack with the use of AI. They want to be in control and avoid sensitive data leakages, misuse of the tool, or brand reputational damage. Langwatch analyses AI solutions, evaluates their quality, and prevents AI risks,” explains Draisma. 

“With that, our mission is to make companies feel confident about launching AI products to the public while taking safety, quality, and usefulness into account,” she states. 

Approach to Quality Assurance

At the core of Langwatch’s offering is a comprehensive evaluation criteria designed to assess the quality of AI solutions from multiple perspectives. The platform leverages three aspects:

• User Feedback Analysis: Langwatch employs sentiment analysis and direct user feedback, combined with insights from internal stakeholders, to gauge real-world performance and user experience of AI solutions.
• Comprehensive Evaluation Library: Langwatch provides a comprehensive library of pre-built evaluations, called “Lang-evals,” designed to identify and reduce common errors made by language models. These evaluations analyse inputs, including user queries, prompts, generated responses, and retrieved-context or source documents, to produce pass/fail scores and explanations, empowering businesses to pinpoint and address potential issues.
• Evaluation Criteria: Recognising the unique requirements of each business, the Dutch company offers the flexibility to define custom evaluation criteria tailored to specific organisational needs and objectives.

Safeguarding against misuse and data leaks

One critical challenge faced by businesses while deploying generative AI solutions is the risk of off-topic conversations and sensitive data leakage. 

Langwatch addresses these concerns through advanced AI models capable of real-time detection of off-topic discussions, enabling companies to steer conversations back on track before generating potentially problematic responses. 

“Similar to data leakage, Langwatch’s PII detection can be used to completely block messages with sensitive content, such as credit card numbers and personal phone numbers, from going out unintended,” she adds. 

Mitigating hallucinations and ensuring output quality

Among the most significant risks associated with the misuse of AI tools are hallucinations—the generation of misleading or entirely fabricated responses—and concerns about overall output quality. 

“What’s particularly troublesome is how difficult it can be to detect these fabricated responses! They may even seem plausible at first glance, making them incredibly elusive. The ramifications of these errors are significant—a single instance can leave users feeling uncertain and paint your product as unreliable or untrustworthy. This is why it is essential to make sure the tool is being used for only the intended purpose and measure the quality to be able to reduce hallucinations as much as possible,” she conveys.

Real-world impact

Langwatch is approached by companies in the experimental stage of their “Proof-of-concept” where they manually evaluate the AI responses for quality through eyeballing.

“They are, however, scared about the moment when the magical thing goes into production for tens of thousands of users. The ability to identify when the AI is going off the grid will help improve the LLM solution. Still, it’s also critical to act quickly for the users’ benefit,” explains Draisma. 

Currently, Langwatch is assisting mid-market businesses in developing LLM-powered applications either in-house or for their customers.

“Through this process, we are not only helping them but also learning from the growing number of GenAI startups. These startups require assistance in understanding how their users are utilising their products to achieve their ‘product-market-fit’,” she adds.

Funding 

Langwatch raised its first pre-seed investment via Antler in January. 

In February, the Dutch startup launched its first Minimum Viable Product (MVP) with initial pilot customers, successfully converting them into paying customers by April. 

They also secured funding from Rabobank. 

“With this giant and speed of traction, plus seeing the high need for this product, we are aiming to close another round before summer to focus on the further development of their product fully,” she reveals. 

Antler’s role

Langwatch’s journey has been significantly shaped by the support and guidance provided by Antler. In fact, the early-stage investor played a pivotal role in facilitating the formation of the founding team. 

“While Manouk initially aimed to find the ideal co-founder during her time at Antler, this goal wasn’t met within the program. However, Antler continued to support her efforts to find a suitable partner,” says RJ Schuurs, Partner at Antler to Silicon Canals.

“When Manouk eventually teamed up with Rogerio, Antler quickly recognised the potential of their partnership and made an immediate investment. Both the perseverance of Manouk and this ongoing support and belief in the founder’s vision significantly shaped Langwatch’s path,” he adds, 

Beyond funding, the company benefited from Antler’s extensive network of potential co-founders, connections to other portfolio founders, pitching feedback, and strategic advice on funding strategies, accelerating the resolution of early-stage challenges and propelling the business forward.